Install Six Everyday Agent Skills from Chinese Creator Khazix

Most skill collections chase one domain. khazix-skills, a 21k-star repo from Chinese AI blogger 数字生命卡兹克 (Khazix), is a personal daily toolkit: six skills covering goal-setting, disk cleanup, AI news, project hygiene, research, and writing, each open-sourced only after the author ran it in his own projects.

Why This Skill Matters

The README is primarily Chinese — an English version exists alongside it — and the collection follows the Agent Skills standard, with Claude Code, Codex, Qoder, Kimi Code, iFlow, CodeBuddy, and Cursor listed among the 40+ compatible agents. The six skills divide cleanly:

• leader turns a vague idea into a goal brief your agent can run for hours.
• storage-analyzer scans your whole disk through the agent and grades what to clean.
• aihot pulls a daily AI news digest with no API key.
• neat-freak re-aligns docs, rules files, and agent memory after each session.
• hv-analysis produces long-form research reports on a product, company, or concept.
• khazix-writer writes long articles in the author's own published style.

Installation

The documented route is a single sentence to your agent:

帮我安装这个 skill：https://github.com/KKKKhazix/khazix-skills/tree/main/leader

Swap the directory name for whichever skill you want — the agent clones it into the right place itself. If your agent cannot load skills at all, the README says to download that skill's SKILL.md and use it as a project rules file or paste it into the conversation; the effect is the same.

Real Workflow: Turn a Vague Idea into a Goal Brief

Type one line to trigger leader:

帮我给 agent 写个目标

Leader first verifies the codebase itself — its zeroth rule is that the map must be tested, not heard — then asks you at most five questions only you can decide, and writes a Markdown goal brief in about twelve minutes, per the README. The brief is built on seven questions: purpose, done state, evidence, anti-cheating, boundaries, trade-offs, and unknowns. The output is plain Markdown, so you can paste it into Claude Code's /goal, Codex's goal mode, or any agent with no goal mode at all.

Real Workflow: Clean Your Disk Through the Agent

Say 帮我看看存储 (or C 盘满了) and storage-analyzer runs a read-only scan, then opens an interactive HTML report with a three-color list: green for pure cache that regenerates, yellow for user data where the report offers only open-in-folder and move-to-trash, and red for running applications and system files — explained, with no delete button at all. Deletion requires clicking a button in the browser plus a second confirmation dialog, and the local server binds to 127.0.0.1 with a random port and token. The README notes macOS is fully tested, the Windows code is ready, and first-time Windows users should take care.

Tips

• Install per skill — you only need the directory URL of the one you want, not the whole repo.
• Run /neat at the end of a long session: neat-freak aligns CLAUDE.md/AGENTS.md, project docs, and agent memory with what actually changed, and proposes deletions as candidate lists you confirm.
• hv-analysis runs a vertical line (the subject's history) and a horizontal line (its contemporaries) and typesets the findings into a 10,000–30,000-character PDF report.
• khazix-writer refuses corporate filler — the README lists banned phrases like 赋能 and 闭环 — so expect a voice with opinions, not a neutral ghostwriter.

When Not to Use This

If you do not read Chinese, friction is real: the triggers, the README, and the per-skill documentation are Chinese-first. This is a personal collection, not a platform — there is no unified CLI beyond asking your agent per skill, and no team features. Storage-analyzer, finally, is a disk aid, not a backup: the three-color grading still leaves every yellow-light decision to you.

See the leaderboard for more skills.